Update rocMLIR integration API to version 5 (perfConfig, cluster size) by dhernandez0 · Pull Request #4761 · ROCm/AMDMIGraphX

dhernandez0 · 2026-04-09T09:48:28Z

MERGE AFTER #4748

Motivation

Update MIGraphX integration to support rocMLIR API version 5, which passes perfConfig through the backend pipeline (for triton backend compatibility) and returns cluster size from kernel attributes (for gfx1250+).

rocMLIR PR: ROCm/rocMLIR#2341 (I'll update the commit hash when merged).

Technical Details

Bump MLIR_MIGRAPHX_DIALECT_API_VERSION check from 4 to 5
mlirMIGraphXAddBackendPipeline now takes a MlirMIGraphXBackendOptions struct (arch, perfConfig, optLevel) instead of a separate arch string; the tuning solution is extracted and passed as perfConfig
Update get_launch_params to return cluster size from mlirGetKernelAttrs (currently ignored with a TODO)

Changelog Category

Add a CHANGELOG.md entry for any option other than Not Applicable

- Added: New functionality.
- Changed: Changes to existing functionality.
- Removed: Functionality or support that has been removed. (Compared to a previous release)
- Optimized: Component performance that has been optimized or improved.
- Resolved Issues: Known issues from a previous version that have been resolved.
- Not Applicable: This PR is not to be included in the changelog.

dhernandez0 · 2026-04-09T09:49:06Z

src/targets/gpu/mlir.cpp

+        }
        // 2nd pipeline to call
-        run_backend_pipeline();
+        run_backend_pipeline(solution_str);


I undertsand solution is never null when we call run_backend_pipeline, is that correct?

Yes that should be the case. Probably should throw an exception if its null. Then you can just do: run_backend_pipeline(solution.to<std::string>())

dhernandez0 · 2026-04-09T09:56:55Z

src/targets/gpu/mlir.cpp

+        MlirMIGraphXBackendOptions opts{};
+        opts.arch       = target_arch.c_str();
+        opts.perfConfig = solution.c_str();
+        opts.optLevel   = 3;


added this because it could be useful for debugging on your side as well. It's often the case that LLVM bugs go away if you pass optLevel=0.

dhernandez0 · 2026-04-09T09:59:18Z

I currently have this split into two PRs: #4748 (rock. prefix for attributes) and this one (API v5: perfConfig + cluster size). Happy to squash them into a single PR if you'd prefer, please let me know. @pfultz2

Copilot

Pull request overview

Updates MIGraphX’s rocMLIR integration to match rocMLIR MIGraphX dialect API v5 by switching the backend pipeline hook to the new options-based API and by reading the newly returned kernel cluster size attribute.

Changes:

Bump MLIR_MIGRAPHX_DIALECT_API_VERSION compatibility check from 4 → 5.
Update backend pipeline invocation to use MlirMIGraphXBackendOptions (arch, perfConfig, optLevel).
Extend get_launch_params() to return cluster size (currently ignored by code_object_op).

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-04-09T10:05:57Z

src/targets/gpu/mlir.cpp

-        mlirMIGraphXAddBackendPipeline(pm_back.get(), target_arch.c_str());
+        MlirMIGraphXBackendOptions opts{};
+        opts.arch       = target_arch.c_str();
+        opts.perfConfig = solution.c_str();


opts.perfConfig is always set to solution.c_str(). When solution is empty (common path when no explicit tuning solution is provided), passing an empty string may be interpreted as an explicit perfConfig by rocMLIR and could cause parse/validation failures. Prefer setting perfConfig to nullptr when solution.empty() (or only setting it when a solution exists).

Suggested change

opts.perfConfig = solution.c_str();

opts.perfConfig = solution.empty() ? nullptr : solution.c_str();

Copilot · 2026-04-09T10:05:58Z

src/targets/gpu/mlir.cpp

+        MlirMIGraphXBackendOptions opts{};
+        opts.arch       = target_arch.c_str();
+        opts.perfConfig = solution.c_str();
+        opts.optLevel   = 3;
+        mlirMIGraphXAddBackendPipeline(pm_back.get(), &opts);


opts.optLevel is hard-coded to 3 without explanation or configuration. If MIGraphX supports varying optimization levels elsewhere, consider threading that setting through (env var / compile option) or at least defining a named constant to avoid an unexplained magic number here.

Copilot · 2026-04-09T10:05:58Z

src/targets/gpu/mlir.cpp

            set_tuning(solution);
+            solution_str = *solution.if_string();


After set_tuning(solution) validates that solution is a string, the code re-fetches it via solution.if_string() and dereferences again to populate solution_str. Consider reusing the validated pointer (or have set_tuning return the string) to avoid redundant lookups and make the control flow clearer.

Suggested change

set_tuning(solution);

solution_str = *solution.if_string();

const auto solution_ptr = solution.if_string();

set_tuning(solution);

solution_str = *solution_ptr;

Copilot · 2026-04-09T10:05:58Z

src/targets/gpu/mlir.cpp

+#if !defined(MLIR_MIGRAPHX_DIALECT_API_VERSION) || MLIR_MIGRAPHX_DIALECT_API_VERSION != 5
 #warning "Incompatible version of rocMLIR library used, disabling"


PR metadata indicates this change should be included in CHANGELOG.md (category marked "Added"), but this PR does not include a changelog entry. Please add an appropriate CHANGELOG.md entry (or update the PR category to "Not Applicable" if that was intended).

dhernandez0 and others added 3 commits April 7, 2026 17:35

Add rock prefix for rocMLIR attributes

fa6ce8a

Merge branch 'develop' into add_rock_prefix

487d134

Update API with rocMLIR to support cluster launch and triton backend

a991bd2

dhernandez0 self-assigned this Apr 9, 2026

dhernandez0 requested a review from causten as a code owner April 9, 2026 09:48

dhernandez0 commented Apr 9, 2026

View reviewed changes

dhernandez0 changed the title ~~Update API with rocMLIR to support cluster launch and triton backend~~ Update rocMLIR integration API to version 5 (perfConfig, cluster size) Apr 9, 2026

fix

3033ef0

dhernandez0 commented Apr 9, 2026

View reviewed changes

fix

9f30552

dhernandez0 requested review from CharlieL7, anisha-amd, bdevorem, Copilot, pfultz2 and umangyadav April 9, 2026 10:00

Copilot started reviewing on behalf of dhernandez0 April 9, 2026 10:02 View session

Copilot AI reviewed Apr 9, 2026

View reviewed changes

Addressing PR comments

84b877e

pfultz2 approved these changes Apr 10, 2026

View reviewed changes

dhernandez0 force-pushed the add_rock_prefix branch from d14c41b to 5e0001d Compare April 11, 2026 04:32

dhernandez0 requested a review from a team as a code owner April 11, 2026 04:32

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Update rocMLIR integration API to version 5 (perfConfig, cluster size)#4761

Update rocMLIR integration API to version 5 (perfConfig, cluster size)#4761
dhernandez0 wants to merge 6 commits intoadd_rock_prefixfrom
rocmlir-integration-cluster-and-triton

dhernandez0 commented Apr 9, 2026 •

edited

Loading

Uh oh!

dhernandez0 Apr 9, 2026

Uh oh!

pfultz2 Apr 9, 2026

Uh oh!

dhernandez0 Apr 10, 2026

Uh oh!

dhernandez0 Apr 9, 2026

Uh oh!

dhernandez0 commented Apr 9, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 9, 2026

Uh oh!

Copilot AI Apr 9, 2026

Uh oh!

Copilot AI Apr 9, 2026

Uh oh!

Copilot AI Apr 9, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

	opts.perfConfig = solution.c_str();
	opts.perfConfig = solution.empty() ? nullptr : solution.c_str();

		#if !defined(MLIR_MIGRAPHX_DIALECT_API_VERSION) \|\| MLIR_MIGRAPHX_DIALECT_API_VERSION != 5
		#warning "Incompatible version of rocMLIR library used, disabling"

Conversation

dhernandez0 commented Apr 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

MERGE AFTER #4748

Motivation

Technical Details

Changelog Category

Uh oh!

dhernandez0 Apr 9, 2026

Choose a reason for hiding this comment

Uh oh!

pfultz2 Apr 9, 2026

Choose a reason for hiding this comment

Uh oh!

dhernandez0 Apr 10, 2026

Choose a reason for hiding this comment

Uh oh!

dhernandez0 Apr 9, 2026

Choose a reason for hiding this comment

Uh oh!

dhernandez0 commented Apr 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Apr 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 9, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

dhernandez0 commented Apr 9, 2026 •

edited

Loading

dhernandez0 commented Apr 9, 2026 •

edited

Loading